The significant risks to safety and biodiversity posed by railway collisions with humans and animals, which are exacerbated by delayed detection and long stopping distances, continue to be a concern along India’s vast railway network. Existing systems like KAVACH and Gajraj aimed at mitigating this issue, suffer from geographical limitations. This research focuses on the development of a computer visionbased real-time image enhancement and obstacle detection program for deployment on trains. The program can be deployed on a system using a high definition camera and a Raspberry Pi 5 microcomputer for onboard processing of video frames. The video frames captured by the camera are enhanced dynamically, based on the illumination conditions, using CLAHE, Retinex, and gamma correction, followed by object detection, using the YOLO model trained on COCO dataset classes. The system achieved precision values of 0.85–0.89, recall between 0.80–0.87, and F1scores of 0.82–0.88 across day, night, and rainy conditions, with an average processing latency of 80 ms per frame, confirming its real-time applicability on Raspberry Pi 5 hardware. The system effectively improved visibility and reliably detected humans, animals, and vehicles on tracks. The program can be integrated with automatic braking systems in trains and could also pave the way for more widespread deployment across varying types of railroad environments. Future improvements would involve fine-tuning the model for various obstacles, utilizing stereo vision for depth estimation, and developing weather-independent processing, as well as optimizing models for lightweight and edge deployment.